Supermicro rack servers with AMD EPYC and Intel Xeon 6

Twelve fixed Gold Series configurations, from a 16-core 1U for a branch office up to 384 cores in 2U, with four- and eight-server multi-node chassis in between. We deliver them installed and provide the support ourselves.

Supermicro AS-2115GT-HNTR-01-G2AS-2115GT-HNTR-01-G2

Comparison

ModelProcessorMemoryStorageNetworkForm factor
AS-1126HS-TN-01-G22x AMD EPYC 9355 (32C/3.55GHz)1.5TB (24x 64GB DDR5-6400)2x 960GB U.2 NVMe1x dual 25GbE SFP28Rack 1U
AS-2115HS-TNR-01-G21x AMD EPYC 9655P (96C/2.6GHz)1.5TB (24x 64GB DDR5-6400)2x 7.6TB U.2 NVMeNVIDIA BlueField-3 B3220 DPU, 2-port 200GbE / NDR200Rack 2U
AS-2115HS-TNR-02-G21x AMD EPYC 9655P (96C/2.6GHz)768GB (12x 64GB DDR5-6400)2x 7.68TB NVMe1x dual 25GbE SFP28Rack 2U
AS-2126HS-TN-01-G22x AMD EPYC 9965 (192C/2.25GHz)1.5TB (24x 64GB DDR5-6400)2x 960GB M.2 NVMe1x dual 25GbE SFP28Rack 2U
SYS-112H-TN-02-G21x Intel Xeon 6731P (32C/2.5GHz)1TB (16x 64GB DDR5-6400)2x 960GB M.2 NVMe1x quad 25GbE SFP28Rack 1U
SYS-222H-TN-01-G22x Intel Xeon 6740P (48C/2.1GHz)1TB (16x 64GB DDR5-6400)2x 3.8TB U.2 NVMe1x dual 25GbE SFP28Rack 2U
AS-2115GT-HNTR-01-G21x AMD EPYC 9534 (64C/2.45GHz) per node512GB DDR5-4800 per node2x 960GB 2.5" SATA SSD per node1x dual 100GbE QSFP28 per nodeRack 2U, 4 nodes
AS-1115CS-TNR-01-G21x AMD EPYC 9555P (64C/3.2GHz)384GB DDR5-64001x 960GB U.2 NVMe2x NVIDIA ConnectX-7 VPI dual 200G QSFP56 + 1x quad 1GbE RJ45Rack 1U
AS-1115CS-TNR-02-G21x AMD EPYC 9115 (16C/2.6GHz)64GB (2x 32GB) DDR5-64001x 960GB M.2 NVMe1x 2-port 10GbE RJ45Rack 1U
SYS-122C-TN-01-G22x Intel Xeon 6736P (36C/2GHz)1TB DDR5-64002x 3.8TB U.2 NVMe1x dual 25GbE SFP28Rack 1U
AS-3015MR-H8TNR-01-G21x AMD EPYC 4565P (16C/4.3GHz) per node64GB DDR5-5600 per node2x 1.92TB U.2 NVMe per node2-port 10/25GbE AOC per nodeRack 3U, 8 nodes
AS-1015A-MT-01-G21x AMD EPYC 4564P (16C/4.5GHz)32GB DDR5-56001x 1.92TB M.2 NVMe1x dual 1GbERack 1U

How to choose between the twelve rack servers

Start with the ratio of cores to memory, because it decides how many virtual machines fit before one of the two runs out. At one end is the AS-2126HS-TN-01, with 384 cores and 4 GB per core for consolidating many small workloads, and at the other the SYS-112H-TN-02, with 32 GB per core; in between, the AS-1126HS-TN-01 offers 64 fast cores and 1.5 TB for databases and software licensed per core. If the workload lives in memory, the two AS-2115HS-TNR models put 96 cores and up to 1.5 TB under a single processor.

Between 1U and 2U the practical difference is cards. 1U models such as the SYS-122C-TN-01 or the AS-1126HS-TN-01 take less space but do not accept double-width GPUs, whereas 2U models such as the SYS-222H-TN-01 take up to four and hold more bays. As for vendor, EPYC 9005 currently reaches more cores per socket, and Xeon 6 makes sense if your estate is already Intel, because live-migrating virtual machines between nodes requires not mixing vendors, or if you use its built-in accelerators.

When the plan is a cluster, multi-node chassis save space and cabling. The GrandTwin AS-2115GT-HNTR-01 fits four nodes with 64 cores, 512 GB and 100 GbE each into 2U, enough for Proxmox or Kubernetes with Ceph, and the MicroCloud AS-3015MR-H8TNR-01 divides 3U among eight 16-core nodes for dedicated hosting or services that need a machine of their own. In both cases the power supplies are shared by every node, something to keep in mind when designing for availability.

For a branch office, the AS-1015A-MT-01 fits a comms cabinet with its 43 cm depth. The AS-1115CS-TNR-02 is the entry point for an SME that wants to grow without changing the box, with 16 cores on a platform that supports up to 160, and its sibling the AS-1115CS-TNR-01 takes on a completely different role with two 200G ConnectX-7 cards, acting as the metadata node in an AI storage system.

Why buy from us

We deliver the machine installed, and if it ever fails you call the same team that put it into service.

We rack it, configure remote management (BMC) and install RHEL, SUSE, Ubuntu or Proxmox, with whatever runs on top already installed and tested: a Ceph cluster, OpenStack, Kubernetes with the NVIDIA GPU Operator, or Slurm.

When something fails you speak to the people who know that machine because they commissioned it, and if you need it there is 24×7 on-call cover with a contractual SLA. It's the same independent support we provide for IBM Power, Linux and Ceph estates.

If you need different memory, drives or networking, we start from the same base model. Tell us what you need in the custom server configurator and we'll call you back within 24 hours with a proposal.

We are an IBM Business Partner and a Red Hat, SUSE and Canonical partner, which is what counts when the server has to slot into an environment with existing systems and licences.

We work across Europe in all three languages, support included, so if your company has offices in Madrid and Brussels, the same team looks after both.

Frequently asked questions

AMD EPYC or Intel Xeon 6?

In this selection EPYC reaches more cores per server, up to 384 with the two EPYC 9005 in the AS-2126HS-TN-01, and more memory per socket thanks to its twelve channels. The family also mixes generations: EPYC 9004 and 9005 in the data centre servers, and EPYC 4004 and 4005 in the entry-level models and the MicroCloud. Xeon 6 pays off if the rest of your estate is Intel, because live migration of virtual machines does not work across vendors, or if your software uses its built-in accelerators. For most virtualisation workloads either will do, and the decision comes down to what you already have.

When is a multi-node server worth it?

When you are going to build several identical servers into a cluster anyway. A four-node GrandTwin or an eight-node MicroCloud takes up less rack space and less cabling than the same number of standalone servers, at the cost of sharing chassis and power supplies. If you only need one or two servers, a conventional 1U or 2U is simpler.

1U or 2U?

A 1U takes less space and is enough if you are not adding GPUs or many cards; core density is set by the processor, not the form factor. A 2U leaves room for double-width GPUs, more bays and better cooling, and is the prudent option if the workload may change over the next few years. Then again, if the rack is short on watts, saving units does little good.

Which one should I choose for Proxmox or VMware?

VMware vSphere is licensed per core, with a minimum of 16 per processor, so a few fast cores make sense, such as the 32 in the SYS-112H-TN-02. The Proxmox VE subscription is per socket, and there processors with many cores pay off, such as the AS-2126HS-TN-01, or a GrandTwin with Ceph if you want high availability from the start.

Rack or tower server?

If you have a rack cabinet or a data centre, go for rack: it takes less space, cabling is tidier and all of these models are designed for it. If there's no cabinet in the office, look at the mini towers in the business computers family, and if there's a small one, the AS-1015A-MT, a 1U unit just 43 cm deep.

Which one fits what you're building?

Tell us about the workload and the data centre it's going into, and we'll come back with the price, a confirmed lead time and what it would take to get it running.